Emissions Trading Systems, Structure Adjustment and Air Pollution Reduction: Evidence from Enterprises in China

Abstract

Using a Chinese-city-piloted emissions trading system (ETS)’s survey data, this study provides nationwide causal estimates of the reduction effects of the implementation of an ETS on air pollution (AP) at the enterprise level. We employ a multiperiod difference-in-differences model to control for potential endogenous problems. The results indicate that the implementation of an ETS significantly reduces the AP of enterprises in pilot areas by 6.96%, and that the effect has a dynamic effect. Heterogeneity analyses show that the emissions reduction effect of an ETS will differ for various enterprises. In terms of region, the implementation of an ETS significantly reduces the AP of eastern, central, and western enterprises by 11.46%, 5.31%, and 12.37%, respectively; in terms of scale, small- and medium-sized enterprises benefit the most (7.69%), followed by large enterprises (1.73%); and in terms of ownership, private enterprises achieve a 7.27% reduction in AP. Additionally, we find that the AP reduction effect is realized by the adjustment of enterprises’ energy and production structures rather than by adding air sewage equipment. Overall, since China is the largest AP emitter worldwide, this study will not only have important implications for sustainable development in China but also the world, providing a scientific basis for starting pilot policies in other developing countries.

1. Introduction

In recent years, as climate change and the health risks caused by greenhouse gas emissions have become global issues [1,2], how to reduce environmental pollution, especially air pollution (AP), has become the core of global sustainable development. Emerging literature documents market-based instruments (MBIs) in environmental regulations as an effective way to control emissions [3,4,5,6]. Among the many MBIs, an emissions trading system (ETS) is undoubtedly one of the critical means with which to solve environmental risks [7,8]. Ever since an ETS was first proposed [9,10], academic research on its effectiveness in pollution treatment has continually emerged. Though many scholars believe that an ETS can effectively improve environmental pollution and facilitate the use of clean energy [4,8,11], there are still some voices that claim that the emissions reduction effect of an ETS is not significant [5,12]. The reason for this debate is that the boundary demarcation of an ETS pilot is inaccurate.

The debate on ETS emissions reduction is particularly evident in the research on developing countries with an imperfect ETS market [3,4,13]. Taking China, the largest AP emitter in the world (According to the World Energy Outlook 2020, published by the U.S. Energy Information Administration, China is the world’s largest emitter of air pollutants in 2020, accounting for 17.7%, 16%, and 8.8% of the total global emissions of SO₂, NOx, and PM10, respectively), as an example, scholars mostly use provincial data when studying the emissions reduction effect of an ETS; however, an ETS is issued at the city-level by local governments. When the ETS is used to discover the effects of environmental regulations on provincial AP [13], the provincial discharge of SO₂ [3], and the discharge of CO₂ as well as NO_x [4,13], the effect of an ETS policy may be mixed with measurement hybrids from accidental wrong targets. While these estimates from the province level are helpful for evaluating the environmental effects of an ETS or investigating the benefits, their external validity is of concern in evaluating this city-targeted pollution reduction policy.

In addition to an ETS, there is also an extensive strand of literature focusing on the effectiveness of carbon emissions trading, another representative of MBIs. Most scholars have affirmed the positive effects of carbon emissions trading on environmental treatments, especially in improving air quality. For instance, Dong et al. [14] claimed that a carbon emissions trading policy significantly reduces carbon emissions and improves air quality by enhancing the innovation ability of local industries. By summarizing the experience of China’s sustainable development, Huang et al. [15] demonstrated that carbon emissions trading provides a reference for developing countries to reduce carbon emissions and address global climate issues. Nonetheless, plenty of opposing evidence has also appeared. Its opponents hold that a carbon emissions trading policy not only reduces the productivity of enterprises [16,17] but could also induce higher pollution according to the pollution haven hypothesis. Additionally, Sinn [18] argued that the implementation of climate policies implies the formation of a stricter regulation system, which in turn accelerates enterprises’ use of fossil fuels. Although there is no consensus on whether carbon emissions trading will promote air quality improvement, and some scholars have further analyzed the synergistic effect of carbon emissions trading and ETSs on air pollution enhancement, we should clarify the connections and differences between carbon emissions trading and ETSs. For instance, regarding policy objectives, carbon emissions trading mainly targets greenhouse gases, represented by CO₂, while ETSs target pollutants, represented by SO₂.

This paper uses a Chinese-city-piloted ETS’s survey data to make a comprehensive causal estimation of the impact of an ETS on AP from the enterprise level. This paper selects enterprises as the research focus for the following two reasons: First, enterprises are important participants and primary energy consumers [19]. The role of entrepreneurs in economic development and environmental protection has become a hot topic among scholars. For instance, Wang et al. [20] found that entrepreneurship-based bottom-up forces could provide better services and protect the environment more effectively under a free market. Second, the aggregation bias often occurring in the use of macro-data could be minimized if we choose micro-enterprise data instead, improving the accuracy of the estimation of policy effects [21]. The ETS survey includes small- and medium-sized enterprises (SMEs) with total assets of less than RMB 400 million and large enterprises with total assets of more than RMB 400 million. It includes all state-owned and non-state-owned enterprises, rendering evaluations of nationwide environmental policies feasible. For the empirical part of our analysis, the average treatment effect of policies on pollution was −0.0696. Holding other factors constant, the implementation of an ETS will significantly reduce pollution.

Since previous papers have focused on policy effects at the provincial level, their external validity deserves attention. The key assumption of previous studies is that all cities in the same province have implemented this policy at the same time, which does not match the reality [22,23]. For example, Hubei Province announced that the policy would be implemented in March 2009, but Ezhou City was the first city in Hubei that implemented it in March 2010. Cities such as Wuhan, Huangshi, and Suizhou eventually implemented the policy after 2014. Therefore, it would be inaccurate if we were to evaluate the policy in 2007 or 2009 as the unified implementation date for the whole province.

The primary obstacles in estimating the effect of an ETS on the AP of enterprises at the municipal level are how to match the Chinese industrial enterprise pollution database with the Chinese industrial enterprise database as well as how to match our city-level ETS survey data with the matched database [8,21]. Given that industrial enterprise is an important part of the Chinese economy, as well as the fact that testing the policy’s effects on industrial enterprises is also a core concern for policymakers, the most recent pollution data of Chinese industrial enterprises were selected as research objects in this paper. Specifically, we used data on corporation names and legal entities to match dozens of indicators, including operating year, the total value of outputs, and administrative regions, with extensive data provided by the Chinese industrial enterprise database. Next, we use the administrative code of China’s cities to match our survey results with these compiled data. In addition, a GDP deflator was chosen as a proxy for an inflation indicator, aiming to eliminate the price factor from the nominal variables. Finally, we adjust the data according to accounting principles to improve the data quality.

We provide comprehensive, nationwide causal estimates of the effect of an ETS on AP for enterprises in China, encompassing all channels of effects. Using the pollution data of Chinese industrial enterprises to measure pollution, we can consider all of the enterprises in China’s industrial survey. In addition, we collected the exact implementation dates of an ETS in more than 500 cities from 2008 to 2021 through field work (Figure 1). We used a multiperiod DID model to empirically evaluate the AP reduction effect of an ETS. The estimation results show that environmental regulation can promote AP reduction in enterprises. Further analyses indicate that increasing the amount of air sewage equipment is not the method by which enterprises can achieve emission reduction targets. Adjusting the energy structure and production structure is the real method by which enterprises can achieve emissions reduction targets.

We use municipal policy shock as the core independent variable, which differs from the previous literature that used provincial pilot shock as the core independent variable. To verify the scientificity of the method in this paper, we introduced the interaction term $Proreform$ between the dummy variable of the provincial pilot and the shock point in the benchmark model for analysis. The regression results of this paper show that, after introducing provincial policy shock into the model, municipal policy shock still has a significant negative impact on AP, while provincial policy shock does not. This means that taking the impact assessment down to the municipal level can reduce the distraction of other unintended false targets, consistent with the above conjecture in this paper.

Previous papers in this literature have mainly looked at policy effects over a short period [22,24]; however, during the transition period, many institutional reforms in China can initially play a positive role. Over time, a decline in policy execution will significantly reduce the dynamic effect of a policy. In China, an ETS is not yet complete. Considering the time lag in the effectiveness of an ETS, we test the dynamic effect of an ETS on AP reduction. This estimation result is useful. We found that the AP reduction effect of an ETS is not only effective in the short term but can also alleviate the AP of enterprises in the long run.

This article contributes to existing papers in four main ways: First, this paper investigates the impact of the implementation of an ETS on enterprises’ AP from the piloted ETS on a city level. We combine the Chinese industrial enterprise database with the Chinese-city-piloted ETS survey database to establish a comprehensive understanding of policy effects on enterprise environmental performance. Enterprise micro-data allow us to analyze the enterprise-level impact mechanisms and heterogeneous effects of an ETS on different enterprises. Moreover, micro-data do not suffer from the aggregation bias of macro-data, so we can improve the accuracy of the policy evaluation. Second, we introduce both municipal and provincial policy shock variables into the model to analyze whether provincial policy shock as a research pilot is biased. The regression results show that provincial policy shock has no significant effect on enterprises’ AP when two shock variables are introduced into the model simultaneously. Third, we construct a dynamic analysis framework to analyze the long-run effect of an ETS. The main reason for the dynamic analysis in this paper is that, for China in the transition period, many policy reforms could have a positive impact initially; however, as time goes on, the decline in policy execution may greatly reduce the dynamic effect of the policy. Finally, in terms of the influence mechanism, we found an interesting phenomenon. Specifically, enterprises do not increase air sewage equipment in response to the implementation of an ETS. The actual influencing mechanism of an ETS on an enterprise’s AP is the adjustment of the energy structure and production structure of enterprises.

The paper is organized as follows: Section 2 delivers the background of ETS implementation. Section 3 describes the empirical model and the data. Section 4 presents the empirical results and discussions. Section 5 provides a further mechanism analysis. We conclude the study in Section 6.

2. Policy Background and Characteristic Facts

As an environmental means to realize the development of a green economy, an ETS plays an essential role in solving environmental pollution [7]. The ETS scheme was first proposed by the American economists Crocker [9] and Dales [10]. Its primary idea is that the government, on behalf of the public and the owners of environmental resources, allocates emission rights under the total amount limit to polluters, with or without compensation. The polluters can sell or transfer these rights to others, and then they can buy the rights from the government or the polluters who own the rights [25].

The practice of an ETS in a narrow sense in China can be traced back to 1988. China’s State Environmental Protection Administration (CSEPA) promulgated “Interim Measures for the administration of water pollutant discharge permits”, which stipulates that the total amount control index of water pollution can be adjusted among pollutant discharge units. From 1990 to 1994, CSEPA conducted a pilot of the AP permit system in 16 key cities across the country in addition to a pilot of a SO₂ and soot ETS in 6 key cities (included in the above 16 cities). In 1996 and 2000, China’s State Council successively issued the “National total emission control plan of major pollutants during the Ninth Five Year Plan period” and the “Air pollution prevention and control law”. The pollution control policy changed from concentration management to total emissions management, providing legal and policy support for the implementation of emissions trading.

Jiangsu Province is located in the Yangtze River Delta. Its rapid economic growth since the reforms and opening up has led to increasing demand for electricity, and the expansion of old plants as well as the construction of new power plants must be carried out. In the process of expansion and construction, Taicang Harbour Environmental Protection Electric-power Generation Co., Ltd. (Taicang, China) has fully used sulfur dioxide desulfurization devices; however, 1700 tons of sulfur dioxide emissions targets still need to be appropriately addressed. Therefore, to achieve the goal of not breaking the total pollution emissions targets while achieving economic growth, Nantong, Jiangsu Province, China, became the first ETS pilot city in China in September 2001. In March 2002, alongside the American State Environmental Protection Association, CSEPA launched a research project on “Promoting the implementation of China’s total sulfur dioxide emission control and ETS” in Shandong, Shanxi, Jiangsu, Henan, Shanghai, Tianjin, Liuzhou City, and the China Huaneng Group Corporation. This was the largest demonstration of an ETS launched at that time. In 2007, the pilot work of an ETS was officially launched in 11 provinces: Jiangsu, Tianjin, Zhejiang, Hebei, Shanxi, Chongqing, Hubei, Shanxi, Inner Mongolia, Hunan, and Henan. In the same year, Jiaxing, Zhejiang Province, saw the establishment of the nation’s first domestic emission trading center, which signaled the gradual institutionalization, standardization, and globalization of emissions trading in China. In 2014, the general office of the State Council issued “The guiding opinions on further promoting the pilot work of compensated use and trading of emission rights” (2014, No. 38), which shows that ETS use is gradually extending in China. Since 2018, more than 200 cities in China have implemented an ETS (Figure 2).

Since 1968, when an ETS was proposed by Dales [10], the research on ETSs has been endless. Some studies have shown that an ETS has a limited or no effect on emissions [5,12]. Shin [5] found that China’s ETS experiment is not institutionalized in any province, which is essentially invalid. Borghesi et al. [12] found that implementing an ETS in Italy has a limited policy effect due to quota issuance being too loose. Further, Li et al. [26] pointed out that the ETS pilot policy has been ineffective in reducing emissions in some areas of China.

Though the above studies hold negative conclusions, extensive studies support that implementing an ETS can significantly reduce emissions [4,11,13]. For instance, Dai et al. [4] argued that provincial CO₂ and NO_x emissions will be reduced by 33% and 31%, respectively, in an assumed environmental regulation scenario with an ETS in Guangdong Province. Similarly, Zhang et al. [13] pointed out that ETSs have shown a relatively positive side in carbon emissions reduction. It is worth noting that most studies use regional data. In addition, by using provincial data in China, Zhang et al. [13] proved the effectiveness of the pollution reduction effect of ETSs. Furthermore, Zhang et al. [24] provided solid empirical evidence that carbon emissions have been significantly reduced after the implementation of ETSs. Additionally, few studies use industry-level data to investigate the policy effects of ETSs [27]; however, enterprises, the main participants in economic activities and the main consumers of energy [19], have been overlooked in the papers exploring the policy effects of ETSs. Therefore, the key in this paper for measuring effectiveness lies in the responses of enterprises to ETSs. Whether the AP of enterprises, the main emissions reduction target of an ETS, is affected by an ETS remains to be further tested.

As a market-based environmental regulatory instrument and economic institution, an ETS motivates enterprises to reduce pollution. Traditionally, pollutant emissions reduction has mainly relied on administrative orders and entrepreneurs’ social responsibility, which led to poor treatment results; however, after the implementation of an ETS, on the one hand, enterprises that effectively reduce emissions will have a surplus of emission rights, and they could sell or transfer the surplus rights to obtain economic returns, which is the market’s compensation for enterprises’ environmental protection behavior. On the other hand, the buyer needs to pay for the excess emission rights, and the expense is the cost of environmental pollution. The significance of an ETS is that it allows enterprises to increase their motivation to reduce emissions for their own benefit, such that the goal of reducing pollution can be achieved.

Some studies have further explored the route of an ETS to achieve emissions reduction. Several academics contend that an ETS will enhance environmental quality by motivating enterprises to innovate technologically, facilitating the development of low-carbon technologies, promoting the usage of renewable energy, and enhancing energy efficiency [8,28,29]. Additionally, some scholars focus on the impact of an ETS on energy structure, industrial structure, and production structure. For instance, Wu [22] found that the rising transaction price in an ETS can effectively reduce CO₂ by improving the energy structure. Xiao et al. [30] claimed that the reconstruction of production factors such as capital, labor, and land under the promotion of ETS regulation can reduce the use of energy. Promoting energy transformation or the structural adjustment of production factors is a positive response to environmental regulation and another mechanism by which to achieve policy objectives, but few scholars have mentioned it. The theory of sustainable development, a main concern of social development, reflects the contradiction between economic development and environmental protection. For enterprises, emissions reduction might harm their economic interests, but continued pollution will damage the ecological environment. These contradictions promote the adjustment of enterprise energy and production structures so that environmental protection and economic growth can both be achieved.

Specifically, on the one hand, the fossil-fuels-based energy consumption structure is an important cause of air pollution [14]. The implementation of an ETS policy increases the cost of pollutant emissions for enterprises that take coal and oil as their main energy sources. Once such cost exceeds the expense of shifting the energy consumption structure from those that are coal- and oil-dominated to a clean-energy-dominated one; according to the principle of profit maximization, enterprises will reduce their AP emissions through energy structure transformation and thus reduce their costs. On the other hand, an ETS will increase the cost of highly polluting enterprises, which will impact enterprises’ behaviors. The changes in enterprises’ behaviors are directly reflected in the adjustment of the production mode. As we know, the production mode of market outsourcing can make enterprises implement more finely and deal with pollution risks more effectively [31]. Based on the above analyses, this article proposes two hypotheses:

Hypothesis 1.

The implementation of an ETS effectively reduces the AP of enterprises in pilot cities.

Hypothesis 2.

An ETS may lead the energy transformation or structural adjustment of production factors of enterprises to achieve air pollution reduction.

Taking the phased implementation of an ETS in China as a quasi-natural experiment and employing a multiperiod difference-in-differences (DID) model, this paper finds that the implementation of an ETS effectively reduces AP emissions; we additionally introduce energy consumption and production structures into the baseline model for mechanism analyses, through which we find solid evidence supporting Hypothesis 2 and also confirm the correctness of Hypothesis 1.

3. Research Design and Methodology

3.1. Model Design

3.1.1. Multiperiod DID Model

Since the ETS policy was piloted in 2001, it has been gradually implemented in various regions. The implementation of an ETS in China can be regarded as a multiperiod exogenous shock, which provides a proper experimental scenario for this study. The DID model is a widely used method by many scholars to investigate the effectiveness of policy implementation; however, the traditional DID model assumes that all pilot cities implemented the ETS policy at a specific time. Therefore, given an ETS was implemented in phases in China, referring to Yu et al. [32] and Lyu et al. [21], this article employs a multiperiod DID model to investigate the emissions reduction effects of an ETS. The multiperiod DID model is written as follows:

$$A{P}_{ijrt}=\beta \times refor{m}_{i,t}+{\mathsf{\Gamma }}^{\prime }{X}_{i,t}+{\alpha }_i+{\alpha }_j+{\alpha }_r+{\alpha }_t+{\epsilon }_{i,t}$$

(1)

where $A{P}_{ijrt}$, the dependent variable in Model (1), indicates the logarithm of AP emissions of the enterprise $i$ of industry $j$ from city $r$ in year $t$. $refor{m}_{i,t}$, the independent variable of Model (1), is a dummy variable. Specifically, if the located city of enterprise $i$ has implemented an ETS in year $t$, then $refor{m}_{i,t}=1$; otherwise, $refor{m}_{i,t}=0$. ${\alpha }_i$, ${\alpha }_j$, ${\alpha }_r$, ${\alpha }_t$ represent the individual, industry, city, and year fixed effect, respectively. $X_{i,t}$ is a series of control variables, and ${\epsilon }_{i,t}$ is the error term.

Additionally, to verify the rationality of municipal policy shock as the core independent variable in this paper, we introduce another dummy variable, $Proreform$, an interaction term for municipal policy shock and the shock point, into the baseline equation for analysis. The model settings are as follows:

$$A{P}_{ijrt}=\beta \times refor{m}_{i,t}+{\beta }_1\times Prorefor{m}_{i,t}+{\mathsf{\Gamma }}^{\prime }{X}_{i,t}+{\alpha }_i+{\alpha }_j+{\alpha }_r+{\alpha }_t+{\epsilon }_{i,t}$$

(2)

The parallel trend assumption is crucial to the reliability of the DID estimation; however, the trend in the treatment group after treatment is counterfactual, so the effectiveness of the parallel trend cannot be tested by “truth” [33,34]. In this study, we examine the parallel trend hypothesis by using the year of policy implementation as the base year, following the idea of Beck et al. [35]. Specific settings for the test model are as follows:

$$A{P}_{ijrt}={\displaystyle {\sum }_{T=-16}^{-1}{\tilde{\beta }}_{-T}}\times refor{m}_{i,t-T}+{\displaystyle {\sum }_{T=0}^6{\tilde{\beta }}_{+T}}\times refor{m}_{i,t+T}+{\mathsf{\Gamma }}^{\prime }{X}_{i,t}+{\alpha }_i+{\alpha }_j+{\alpha }_r+{\alpha }_t+{\epsilon }_{i,t}$$

(3)

where $refor{m}_{i,t}$ is a dummy variable that symbolizes ETS policy shock. If the located city of enterprise $i$ implements an ETS, $refor{m}_{i,t}$ equals 1, otherwise it equals 0. ${\tilde{\beta }}_{-T}$ and ${\tilde{\beta }}_{+T}$ indicate the policy effects of $T$ periods pre- and post-policy shock, respectively. The observation period in this paper is sixteen years before and six years after the implementation of an ETS. ${\alpha }_i$, ${\alpha }_j$, ${\alpha }_r$, ${\alpha }_t$, $X_{i,t}$, and ${\epsilon }_{i,t}$ are explained in the same way as in Model (1).

3.1.2. Propensity Score Matching (PSM)

This article employs the propensity score matching (PSM) technique to mitigate the endogenous problem due to selection bias. The implementation of the ETS policy can be viewed as a quasi-natural experiment, and thus selection bias is inevitable in the policy implementation process. The PSM method matches each individual in the treatment group to a specific sample in the control group, promising a quasi-natural experiment close to a random experiment. Therefore, referring to Rosenbaum and Rubin [36], this paper uses the PSM method to eliminate selection bias and confounding bias in the process of policy evaluation. The steps are specifically designed as follows:

Step 1: PSM was performed on the treatment samples. The purpose of PSM is to identify members in the control group that share similar characteristics with those in the treatment group to provide counterfactual results:

$$P_i=probit\left(a{p}_i=1|X_i\right)=\mathsf{\Phi }\left(X_i\right)$$

(4)

where $P_i$ is the probability that the enterprise’s city implemented an ETS. $a{p}_i$ is a dummy variable for policy shock. In other words, if the city has implemented an ETS, $a{p}_i$ equals 1, otherwise it equals 0. $\left(\cdot \right)$ is a normal cumulative distribution function. $X_i$ are matching variables that denote the factors affecting the implementation of an ETS. Based on this probability formula, the predicted probability, ${\widehat{P}}_i$, can be obtained.

Step 2: Cities with similar predicted probability values are matched by using the PSM method. We can then obtain a control group of enterprises with similar characteristics to the treatment group. In addition, if city $i$ is a sample from the treatment group, then city $j\left(i\right)$, a similar city matched from the control group, should meet the following requirements:

$$j\left(i\right)=\arg \underset{j}{}\min |{\widehat{P}}_j-{\widehat{P}}_i|,\forall j$$

(5)

3.2. Data Source and Variable Selection

The data of sample enterprises are taken from the Chinese Industrial Enterprise Pollution Emission Database. Referring to Lyu et al. [21] as well as Zhu et al. [37], this article selects the most recent pollution data of industrial enterprises in China (1998–2014) as research objects. Specifically, we use data on corporation names and legal entities to match dozens of indicators, including operating year, total value of outputs, and administrative regions, with extensive data provided by the Chinese industrial enterprise database. Next, we use the administrative codes of China’s cities to match our survey results to this compiled data, eventually obtaining 417,773 observations.

To further enhance the data quality, we additionally make adjustments in accordance with accounting standards. First, the price factor was removed from the nominal variables by using a GDP deflator as a proxy for an inflation indicator. Then, following the idea of Feenstra et al. [38] as well as Zhu et al. [37], samples with the following data characteristics will be removed in this paper: (1) sales revenue below the export value; (2) the value of fixed assets (current assets) being higher than that of total assets; (3) total output value being below or equal to 0; (4) the fixed asset value being below 0; and (5) the starting year being later than the current year or the starting month being later than December. Therefore, the selected samples avoid statistical errors in the accounts and ensure the quality of the empirical analysis.

3.3. Descriptive Statistics

Fuel shortages and AP are the main problems of current as well as future energy and the environment [39]. Regarding reality, this paper, therefore, treats AP as the research object. This article includes the following control variables in the model based on previous literature [24,40,41]: the growth of the gross production value ($growth$), the fixed capital stock ($K$), the profit margin ($profit$), the degree of industrial agglomeration ($Argcro$), industrial upgrading ($indust$), and infrastructure ($infra$). The full definitions and summaries of each important variable are provided in

Table 1. Descriptive statistics of variables.

Variable Symbol	Variable Definition	Frequency	Mean	Std. Deviation	Min.	Max.
$AP$	Log of air pollutant emissions of enterprises	386,777	10.0397	2.2399	0	21.5026
$reform$	ETS pilot area and after the policy time point	534,973	0.1045	0.3059	0	1
$growth$	Growth rate of enterprises’ gross output value	534,973	0.0291	0.2881	−13.1493	6.9265
$K$	Fixed capital stock (absolute amount)	534,973	0.0354	0.2976	0	48.6610
$profit$	Profit margin (output value/RMB 100)	534,973	0.0000	0.0298	−21.5740	0.7164
$Argcro$	Degree of industrial agglomeration (log of location entropy)	534,973	−9.8435	3.2918	−34.3724	6.9694
$indust$	Industrial upgrading (proportion of added value of secondary and tertiary industries)	534,973	83.5494	23.2073	0	99.9700
$\mathrm{infra}$	Infrastructure (log of railway passenger volume)	534,973	8.1249	2.7374	0	11.7550

.

Table 1 further reports the descriptive statistics of all of the variables. The first column of Table 1 shows all of the variable symbols in this paper. For example, the symbol $reform$ represents ETS policy shock. The variable definition then details the calculations and economic meanings of each variable. In addition, we performed a frequency analysis, concentration trend analysis, and dispersion analysis on the data, which are presented in the last five columns of Table 1. For instance, the mean (std. deviation) of AP is 10.0397 (2.2399), indicating that the emission of air pollutants by enterprises is a common phenomenon. The mean value of $reform$ is 0.1045, indicating that the sample size of the treatment group is about 1/10 of all samples after implementing an ETS. The descriptive statistics of enterprises’ characteristics, such as fixed capital stock, are in line with those of relevant studies [42].

4. Empirical Analysis

4.1. Preliminary Observation of Data

As a common choice in policy studies, the DID method provides a policy’s effect through comparing a treatment group’s pre- and post-policy change in comparison with the control group [43]; however, an issue will result from this. When comparing the treatment group with the control group, subjective selectivity will unavoidably appear [44]. In other words, the similarity between the treatment and control groups is the basis for this approach. Otherwise, the DID model used in this paper cannot meet the “randomized experiment” requirement. Thus, our regression results of the model are not causal effects of the pilot policy. Therefore, we first estimate the distribution density of the AP of two groups; the results are shown in Figure 3. It is observed that the density distribution of the AP in the treatment group is very similar to that of the control group. This indicates that samples of both the treatment and control groups have similar characteristics in terms of distribution.

Furthermore, we conduct a trend analysis on the AP in the two groups before the establishment of an ETS based on the methodology of Chipman et al. [45]. Blue dots indicate the AP of treatment group enterprises, while red dots refer to that of control group enterprises. The results in Figure 4 show that samples in the two groups have almost the same trend.

4.2. Policy Effects Estimation

Firstly, based on Equation (1), the baseline regression results are presented in

Table 2. Results of multiperiod difference-in-differences.

Independent Variable	(1)	(2)
$reform$	−0.0738 ***	−0.0696 ***
	(0.0154)	(0.0155)
$growth$		0.0755 ***
		(0.0061)
$K$		0.0776 ***
		(0.0329)
$profit$		−0.0088
		(0.0089)
$Argcro$		0.0059 ***
		(0.0010)
$indust$		0.0010 ***
		(0.0004)
$\mathrm{infra}$		−0.0163 ***
		(0.0040)
$YearFE$	Yes	Yes
$IndustryFE$	Yes	Yes
$CityFE$	Yes	Yes
$EnterpriseFE$	Yes	Yes
$R^2$	0.0334	0.0345
$N$	385,873	385,873

Notes: T-value adjusted by the White covariance matrix in parentheses. *** $p<0.01$.

. The results show that the coefficient of $reform$ passes the significance test in both columns (1) and (2). In particular, the results in columns (1) and (2) demonstrate the impact of an ETS on AP emissions excluding and including control variables, respectively. For instance, after containing all of the control variables, the coefficient of policy shock ($reform$) in our baseline model is −0.0696 ($p<0.01$), indicating that the pilot policy will reduce AP emissions by 6.96% for enterprises in the treatment group compared to those in the control group.

The empirical result is consistent with our hypothesis that the implementation of an ETS could significantly inhibit the AP emissions of enterprises. On the one hand, implementing an ETS will increase the cost of AP [46], directly inhibiting AP. On the other hand, before an ETS is implemented, some high-polluting enterprises will use bribery to reduce enterprise costs. This behavior is a two-way activity between government employees and enterprises. In cases where public interest and environmental quality are closely related, rent-seeking behavior in the environmental management process could lead to the actual amount of emissions of enterprises exceeding environmental capacity or the total amount of emissions set by governments, thus reducing environmental quality and affecting public interest [47,48]. An ETS, as a market-oriented instrument [49], could thus reduce emissions of pollutants by reducing rent-seeking behavior and enabling enterprises to spend their energy and costs on pollution control rather than rent-seeking. Our results, therefore, confirm Hypothesis 1 of this paper: that an ETS can significantly reduce pollutant emissions.

When we introduced the independent variable representing provincial policy shock into the regression model, we found no significant correlation between it and AP. The results are shown in

Table 3. Comparing the results of municipal pilots and provincial pilots.

Independent Variable	(1)	(2)
$reform$	−0.0716 ***	−0.0682 ***
	(0.0153)	(0.0153)
$Proreform$	−0.0128	−0.0094
	(0.0118)	(0.0118)
$Cont$	No	Yes
$YearFE$	Yes	Yes
$IndustryFE$	Yes	Yes
$CityFE$	Yes	Yes
$EnterpriseFE$	Yes	Yes
$R^2$	0.0334	0.0346
$N$	385,873	385,873

Notes: $\mathrm{Cont}$ denotes control variables, the same below. *** $p<0.01$.

. The regression coefficient of the municipal policy shock variable is −0.0682 ($p<0.01$); the regression coefficient of the provincial policy shock variable is −0.0094, which is significantly smaller than the municipal policy shock variable and not statistically significant. As we mentioned in Section 1, although many scholars have discussed the emissions reduction effects of an ETS by using provincial data, ETSs are issued at the city level by local governments in practice. Therefore, only some cities in the pilot provinces have implemented ETSs at the same time. This is a proper reason why the coefficient of municipal policy shock is not significant. Therefore, it is more reasonable and scientific to explore the impacts of ETSs on AP at the city level.

4.3. Robustness Test

While the findings in Table 2 demonstrate that the implementation of an ETS could greatly lower the AP emissions of China’s enterprises, some endogenous issues brought about by missing variables, outliers, and measurement errors cannot be solved. Therefore, this section offers a series of robust tests with which to support our findings.

4.3.1. Parallel Trend Test

In this study, the observation period spans 16 years before and 6 years after policy shock; thus, there are 22 dummy variables in each model. Additionally, this paper also acknowledges that some enterprises will move to other cities during the observation period because an ETS initially raises enterprise costs. Thus, referring to Lyu et al. [21], we introduce city and year interaction fixed effects into the parallel trend test model. The result of the parallel trend test is demonstrated in Figure 5.

The results in Figure 5 show that, before implementing an ETS, all figures of ${\tilde{\beta }}_t$ are insignificant or significantly positive. After the establishment of urban agglomeration, the coefficient of $refor{m}_{i,t}$ shows an opposite trend, showing that ${\tilde{\beta }}_t$ is negative. The opposite trends before and after policy shock support the parallel trend assumption of the DID method adopted in this paper.

4.3.2. Parameter Substitution Test

Columns (1) and (2) in

Table 4. Results of the parameter substitution test.

Independent Variable	(1)	(2)
	SO2 Emissions	NOx Emissions
$reform$	−0.0597 ***	−0.0714 ***
	(0.0187)	(0.0152)
$growth$	0.0758 ***	0.0383 ***
	(0.0063)	(0.0084)
$K$	0.0566 *	0.0547
	(0.0336)	(0.0366)
$profit$	0.0073	0.0118
	(0.0128)	(0.0149)
$Argcro$	0.0062 ***	−0.0002
	(0.0011)	(0.0015)
$indust$	0.0017 ***	0.0001
	(0.0004)	(0.0011)
$\mathrm{infra}$	−0.0154 ***	−0.0744 *
	(0.0041)	(0.0440)
$YearFE$	Yes	Yes
$IndustryFE$	Yes	Yes
$CityFE$	Yes	Yes
$EnterpriseFE$	Yes	Yes
$R^2$	0.0131	0.0390
$N$	378,918	199,655

Notes: *** $p<0.01$, * $p<0.1$.

are the regression results using SO₂ and NO_x as emissions indicators, respectively. All of the regression results were established under the control of enterprise, industry, city, and year fixed effects.

From Table 4, the estimation results are still significantly negative with alternative measures of air pollutants. Based on the pollutant categories listed by China’s 13th Five-Year Plan, the ETS policy lists chemical oxygen demand, ammonia nitrogen, sulfur dioxide, and nitrogen oxide as being within the scope of pollutants. The implementation of an ETS will result in a strict environment supervision system for enterprises. Therefore, it will reduce emissions of pollutant gases within the list of an ETS. In addition, enterprises will gradually shift their energy and production structures in response to the pilot policy and maintaining economic development. Therefore, the implementation of an ETS is not only effective for the aggregated emissions of air pollutants but also effective for individual air pollutants.

4.3.3. PSM-DID

In addition to the parallel trend assumption mentioned before, another essential premise of the DID method is that the treatment groups are randomly selected. This premise indicates no systematic difference in the changing trend in AP emissions among enterprises located in piloted cities and other enterprises over time. Additionally, the difference in AP emissions between enterprises observed by empirical results is not due to a real policy treatment effect but caused by initial differences. Therefore, referring to the methodology of Heckman [50], we employed the PSM method to alleviate the confounding bias and selective bias. We re-estimated Equation (1) by using the data matched by Equations (4) and (5). The regression results are presented in

Table 5. Results of PSM-DID.

Independent Variable	(1)	(2)
$reform$	−0.0929 ***	−0.0904 ***
	(0.0298)	(0.0298)
$Cont$	No	Yes
$YearFE$	Yes	Yes
$IndustryFE$	Yes	Yes
$CityFE$	Yes	Yes
$EnterpriseFE$	Yes	Yes
$R^2$	0.0329	0.0338
$N$	361,497	361,497

Notes: *** $p<0.01$.

.

Columns (1) and (2) in Table 5 show that the estimated coefficient is still significantly negative with or without control variables; the magnitude is even greater than the one of the baseline results. Specifically, after the implementation of an ETS, the AP of enterprises in the pilot area decreased by 9.29 percent every year without control variables and 9.04 percent every year with control variables. Compared with columns (1) and (2) in Table 2, the estimated coefficients of an ETS after matching are about 1.91 percent and 2.08 percent higher, respectively, indicating that the confounding bias and selective bias in this paper will lead us to underestimate the AP reduction effects of the policy.

4.4. Dynamic Effect Test

To explore whether the implementation of an ETS has a dynamic effect on reducing AP, $reform\times \mu$ is introduced into Equation (1). $\mu$ represents the dummy variable of lag time, which is taken as 1 if $\mu$ is three years after the implementation of the policy, otherwise it is 0. The regression results are reported in

Table 6. Results of the dynamic effect test.

Independent Variable	(1)	(2)
	No Time Trend	Including Time Trend
$reform$	−0.0738 ***	−0.0553 ***
	(0.0154)	(0.0151)
$reform\times \mu$		−0.1651 ***
		(0.0215)
$Cont$	Yes	Yes
$YearFE$	Yes	Yes
$IndustryFE$	Yes	Yes
$CityFE$	Yes	Yes
$EnterpriseFE$	Yes	Yes
$R^2$	0.0334	0.0349
$N$	385,873	385,873

Notes: *** $p<0.01$.

. The results show that the implementation of an ETS has a short-term effect and a significantly long-term effect on reducing the AP of enterprises. The findings provide an important direction for exploring environmental regulation to realize the double dividend of the economy and the environment, which provides an important implication for the Chinese government.

4.5. Who Benefits from an ETS?

Unbalanced development is a problem that has troubled China’s economic development for a long time [51]. The effects of implementing an ETS may vary according to the ownership, location, and scale of enterprises.

Table 7. Heterogeneity of regions.

	Region of Enterprise
Independent Variable	(1)	(2)	(3)
	East	Central	Western
$reform$	−0.1146 ***	−0.0531 **	−0.1237 ***
	(0.0224)	(0.0257)	(0.0418)
$Cont$	Yes	Yes	Yes
$YearFE$	Yes	Yes	Yes
$IndustryFE$	Yes	Yes	Yes
$CityFE$	Yes	Yes	Yes
$EnterpriseFE$	Yes	Yes	Yes
$R^2$	0.0485	0.5950	0.0409
$N$	135,488	169,513	80,089

Notes: *** $p<0.01$, ** $p<0.05$.

divides the sample into the East, Central, and Western regions of China, while

Table 8. Heterogeneity of scale and ownership.

	Scale of Enterprise		Ownership of Enterprise
Independent Variable	(1)	(2)	(3)	(4)	(5)
	Small- and Medium-Sized	Large	State-Owned	Collective	Private
$reform$	−0.0769 ***	−0.0173 ***	−0.0573	−0.0721	−0.0727 ***
	(0.0171)	(0.0345)	(0.0518)	(0.0849)	(0.0164)
$Cont$	Yes	Yes	Yes	Yes	Yes
$YearFE$	Yes	Yes	Yes	Yes	Yes
$IndustryFE$	Yes	Yes	Yes	Yes	Yes
$CityFE$	Yes	Yes	Yes	Yes	Yes
$EnterpriseFE$	Yes	Yes	Yes	Yes	Yes
$R^2$	0.0332	0.0513	0.0464	0.0351	0.0349
$N$	335,273	50,600	76,009	17,026	292,838

Notes: *** $p<0.01$.

divides the sample into SMEs and large enterprises according to scale and splits the sample into state-owned enterprises (SOEs), collective enterprises, and private enterprises according to ownership.

The regression results of regional heterogeneity are listed in Table 7. Columns (1)–(3) represent the results for the East, Central, and Western regions of China, respectively. It is clearly presented that an ETS policy significantly reduces an enterprise’s AP emissions in all regions of China, and the effects are even larger in the East and Western regions. Due to the geographical and initial resource endowment advantages, the East coastal region pulls ahead of the Central region in the development of the level of market liberalization. In addition, the solid economic foundation, high education level, and sufficient human resources of East China have attracted many entrepreneurs from around the world. As mentioned in Section 1, entrepreneurship-based bottom-up forces could provide better service and protect the environment more effectively under a free market [20]. Therefore, it is reasonable that an ETS policy could effectively reduce the AP emissions in eastern parts of China.

The regression results of enterprise heterogeneity are demonstrated in Table 8. Columns (1)–(5) of Table 8 represent the results of SMEs (with total assets less than RMB 400 million as the boundary), large enterprises, SOEs, collective enterprises, and private enterprises, respectively.

Columns (1) and (2) in Table 8 indicate that the implementation of an ETS has a significant emissions reduction effect on enterprises of all scales, and the effect is more pronounced for SMEs than for large enterprises. The production process of large enterprises is more transparent than that of SMEs [52,53], which is more conducive to the supervision of regulatory authorities and the measurement of pollution. Even before the implementation of an ETS, large enterprises already own more mature pollution treatment equipment and technology than SMEs [54].

Columns (3)–(5) in Table 8 show that implementing an ETS can significantly promote the reduction in emissions of private enterprises, but the impacts on state-owned and collective enterprises are not statistically significant. Some scholars argue that SOEs will outperform private enterprises in terms of business performance and environmental governance due to their greater sense of social responsibility [55,56]; however, opponents demonstrate that the average productivity of Chinese SOEs is lower than that of private enterprises [57]. According to the latest data from the National Bureau of Statistics, from January to November in 2020, in terms of profit growth, SOEs suffered a 4.9% year-on-year decline, while private enterprises achieved a 1.8% increase. In terms of profit share, SOEs and private enterprises accounted for 16.3% and 19.6%, respectively. It can be seen that SOEs do not perform as well as private enterprises in terms of business performance.

In addition, another strand of literature investigates the effect of corporate ownership on the effectiveness of environmental regulation. Li et al. [58] argue that political connections and an enterprise’s ownership property are important influences on corporate environmental performance. In China in particular, there are still many problems with SOEs, such as the natural link with the government and the lack of property rights [59,60]. Since SOEs are directly controlled by the central or local governments, they have significant advantages in resource allocation, especially financial support [61]. Therefore, SOEs are not sensitive to the compliance cost pressure brought about by an emissions trading system and the economic innovation incentives provided by the sale of emission rights. On the contrary, a market-oriented institutional environment provides a good platform for technological innovation and restructuring for private enterprises [62]. Unlike SOEs, private enterprises, which do not have close ties with the government and uphold the principle of profit maximization, will flexibly adjust their own energy and production structures through equipment improvement and technological innovation to cope with the emissions reduction requirements after the implementation of an ETS. Similarly, compared with private enterprises, collective enterprises are not sensitive to the costs and incentives brought about by the implementation of an ETS.

5. Influencing Mechanism Analysis

The above findings show that an ETS pilot policy can significantly reduce the AP of enterprises. Furthermore, we want to know through which channel an ETS policy affects AP emissions. Ex-post control measures, such as the number of pollutant treatment facilities, are a straightforward and effective way to reduce pollution [8]. Therefore, it is of concern as to whether an ETS affects the number of facilities for treating pollutants. The regression results in column (1) of

Table 9. Results of mechanism analysis.

Independent Variable	(1)	(2)	(3)	(4)
	Number of Facilities	Consumption of Coal	Consumption of Oil	$\mathbf{Production}\mathbf{Structure}\left(\mathit{K}/\mathit{L}\right)$
$reform$	−0.0097	−0.1984 ***	−0.0589 ***	0.0694 ***
	(0.0135)	(0.0294)	(0.0223)	(0.0093)
$Cont$	Yes	Yes	Yes	Yes
$YearFE$	Yes	Yes	Yes	Yes
$IndustryFE$	Yes	Yes	Yes	Yes
$CityFE$	Yes	Yes	Yes	Yes
$EnterpriseFE$	Yes	Yes	Yes	Yes
$R^2$	0.0095	0.0348	0.0172	0.0450
$N$	242,278	385,873	216,487	526,052

Notes: *** $p<0.01$.

show that the number of facilities for treating pollutants did not significantly increase after policy shock, implying that the AP reduction effect is not achieved through the enhancement of facilities. Since SMEs are reluctant to bear the high costs of facilities, this result is consistent with our findings in the heterogeneity test.

As we all know, when enterprises face the pressure of reducing emissions, reducing the use of polluting energy is a way to achieve quick results without extra costs [8,21]. We carried out a regression analysis on enterprises’ coal and oil consumption, the results of which are represented in columns (2) and (3) of Table 9. The results show that, after the implementation of an ETS, enterprises decreased their consumption of coal and oil, indicating that an ETS pilot policy could motivate enterprises to use less polluting energy rather than increase air sewage equipment. The adjustment of an energy structure is one of the main ways through which to reduce the AP of enterprises.

With the continuous development of the economy, technology, and productivity, the economic and market structures are also changing constantly. The energy consumption structures of enterprises are bound to change to adapt to the new developments of the economy and society. With the aggravation of environmental pollution and the emergence of an energy crisis, reducing investment in traditional high-polluting energy is not only required for dealing with environmental regulations but also a benign choice for maintaining the sustainable and healthy development of enterprises.

The deep-seated reason for the structural change discussed above is the transition of factor endowment [63,64]. In China, the essence of the disappearance of labor dividends is the transition of factor endowment. To deal with this change, enterprises must adjust the production structure to adapt to the current factor endowment. Based on this, this paper explores the capital–labor ratio of enterprises, and the results are listed in column (4). The results show that the implementation of environmental regulation can significantly improve the capital–labor ratio of enterprises, which also means a change in the production structure. Therefore, the correctness of Hypothesis 2 could be supported.

6. Conclusions

Taking the progressive implementation of an ETS in China as a quasi-natural experiment and using a large-scale micro-dataset of Chinese industrial enterprises, this paper employs a multiperiod DID method and a PSM-DID model to evaluate the effects of the implementation of an ETS on AP reduction. We also explore the heterogeneity effects in terms of the scale, region, and ownership of enterprises. Furthermore, we clarify the influencing mechanism through which an ETS pilot policy could reduce AP. The results show that the phased pilot policy could significantly suppress the AP of Chinese enterprises. This reduction effect is greater for enterprises in the Western region, followed by those in the East and Central regions. In addition, the policy effect is apparent on large enterprises but insignificant on SMEs. Similarly, private enterprises respond apparently to policy shock, while there is no significant AP reduction effect for state-owned and collective enterprises. Moreover, through the mechanism analysis, the adjustments of enterprises’ energy and production structures are proven to be effective mechanisms by which an ETS policy reduces AP.

Our study has both practical and theoretical implications. From a practical point of view, extensive energy consumption has led to severe global warming in recent years, which will not only hinder the sustainable development of countries worldwide but also harm human survival. China, the largest developing country in the world, is also the largest energy consumer and AP emitter. According to statistics, China accounted for 17.7% of global SO₂ emissions in 2020; therefore, how to use environmental regulations to reduce AP in China will be essential for the sustainable development of China and the world. In addition, it also has implications for developing countries that also face high energy consumption and high air pollution. From a theoretical point of view, whether environmental regulation can promote enterprises’ to reduce emissions has long been a hot topic in environmental economics and science, and an ETS is an environmental regulation instruments. Our results confirm that there is indeed a significant causal relationship between the two, and also shed new light on the debate as to whether an ETS positively or negatively affects AP. Historically, this debate has been about how compliance costs from environmental regulations affect enterprises’ willingness to reduce emissions. Our results confirm that another channel is influencing this debate. An ETS, implemented to reduce AP, will affect enterprises’ energy and production structures, which at least partially offsets the loss of productivity due to compliance.

Based on the main conclusions above, we propose the following policy implications for reducing AP via an ETS pilot policy: First, actively promote the expansion of an ETS pilot policy in China. Moreover, encourage developing countries that also face sever AP to initiate an ETS policy, since this paper has provided a scientific basis for starting pilot policies in other developing countries. Second, enhance the market-oriented reforms of SOEs. There is no doubt that SOEs are the core of the national economy, while low efficiency and high pollution have always been problems for SOEs. The natural linkage with the government causes SOEs to lack an incentive to innovate. It is under a free market that entrepreneurship-based bottom-up forces could protect the environment more effectively [20], and enterprises will have more incentives to innovate to improve their energy and production structures. As a result, they could increase productivity, reduce pollution emissions, and develop a competitive advantage. Third, local governments can introduce related policies to promote the transformation of the energy and production structures of enterprises. The transition of energy and production structures will not only increase the productivity of enterprises, which will improve the economic development of local governments, but also prove to be an important mechanism through which an ETS can exert its ecological governance effects. It provides us with a fresh idea to protect local economic development and reduce AP emissions.